Semantic Search System with Multi-Granularity Text Object Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing search engine systems fail to provide meaningful interpretations of search results, do not enable users to understand why results are relevant, and do not consider the semantic meaning of search queries.
Innovation Solution
A method and system for interactive searching based on semantic similarity, where a trained model generates and compares semantic representations of text objects to determine similarity scores, allowing users to explore similarities at various levels (document, paragraph, sentence, word) and filter results based on calculated attributes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If traditional keyword matching algorithms are used to retrieve search results, then the search process is simple and fast, but the system cannot provide interpretation of why results are relevant and does not understand the semantic meaning of queries
Solution Approach 1:
The patent introduces semantic representations as an intermediary layer between the search query and the document corpus. Instead of directly matching keywords, the system converts both queries and documents into semantic representations (e.g., vector embeddings) that capture their meaning. This intermediary enables the system to understand semantic similarity while maintaining a structured processing pipeline that balances complexity and interpretability.
Solution Approach 2:
The patent replaces traditional mechanical keyword matching algorithms with machine learning-based semantic representation models. These models automatically learn to represent text in a way that captures meaning, substituting the manual design of keyword matching rules with data-driven semantic understanding. This substitution enables the system to provide meaningful interpretations of search results based on semantic similarity scores.
2Loss of information
If semantic representations are generated and compared to determine similarity scores, then the system can provide meaningful interpretations of search results, but the computational complexity and processing time increase
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing semantic representations of documents in the corpus before search queries are submitted. When a query arrives, the system only needs to generate the query's semantic representation and compare it against the pre-computed document representations, significantly reducing processing time. This pre-processing step enables fast retrieval while maintaining semantic understanding.
Solution Approach 2:
The patent segments the search process into distinct stages: query semantic representation generation, document semantic representation retrieval, similarity score computation, and result ranking. This segmentation allows each component to be optimized independently and enables parallel processing of multiple queries against the pre-computed document representations, reducing overall processing time while providing detailed interpretation of results.
3Adaptability or versatility
If the system returns only documents deemed most relevant to a query, then the search results are concise, but the system does not provide other types of connections such as similar sentences or paragraphs
Solution Approach 1:
The patent applies local quality by enabling users to specify different levels of granularity for search results (document level, paragraph level, sentence level, or word level). Instead of returning only full documents, the system can return semantically similar excerpts at the appropriate level of detail. This allows the system to provide diverse types of connections (similar sentences, paragraphs, or documents) while keeping the result quantity manageable by focusing on the most relevant local portions of texts.
Solution Approach 2:
The patent introduces dynamics by making the search result format adaptable based on user needs and query characteristics. The system can dynamically adjust whether to return documents, paragraphs, sentences, or phrases, and can provide different levels of semantic similarity information. This dynamic adaptability enables the system to provide multiple types of connections without overwhelming the user with excessive results, as the quantity and type of results can be adjusted based on the specific search context.
Data Source
AI summary
There is provided a method and a system for generating a graphical representation. A search request comprising an indication of a text object is received and a semantic representation of the search request is generated. One or more documents each satisfying a similarity condition with respect to the semantic representation of the search request are obtained, followed by the identification of one or more of: a respective set of paragraphs that each satisfies the similarity condition; a respective set of sentences that each satisfies the similarity condition; and a respective set of words that each satisfies the similarity condition. Then a graphical representation of at least a portion of each one of: the respective set of paragraphs, the respective set of sentences, and the respective set of words is generated.